2 * Xen leaves the responsibility for maintaining p2m mappings to the
3 * guests themselves, but it must also access and update the p2m array
4 * during suspend/resume when all the pages are reallocated.
6 * The p2m table is logically a flat array, but we implement it as a
7 * three-level tree to allow the address space to be sparse.
13 * p2m_mid p2m_mid p2m_mid_mfn p2m_mid_mfn
15 * p2m p2m p2m p2m p2m p2m p2m ...
17 * The p2m_mid_mfn pages are mapped by p2m_top_mfn_p.
19 * The p2m_top and p2m_top_mfn levels are limited to 1 page, so the
20 * maximum representable pseudo-physical address space is:
21 * P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE pages
23 * P2M_PER_PAGE depends on the architecture, as a mfn is always
24 * unsigned long (8 bytes on 64-bit, 4 bytes on 32), leading to
25 * 512 and 1024 entries respectively.
27 * In short, these structures contain the Machine Frame Number (MFN) of the PFN.
29 * However not all entries are filled with MFNs. Specifically for all other
30 * leaf entries, or for the top root, or middle one, for which there is a void
31 * entry, we assume it is "missing". So (for example)
32 * pfn_to_mfn(0x90909090)=INVALID_P2M_ENTRY.
34 * We also have the possibility of setting 1-1 mappings on certain regions, so
36 * pfn_to_mfn(0xc0000)=0xc0000
38 * The benefit of this is, that we can assume for non-RAM regions (think
39 * PCI BARs, or ACPI spaces), we can create mappings easily b/c we
40 * get the PFN value to match the MFN.
42 * For this to work efficiently we have one new page p2m_identity and
43 * allocate (via reserved_brk) any other pages we need to cover the sides
44 * (1GB or 4MB boundary violations). All entries in p2m_identity are set to
45 * INVALID_P2M_ENTRY type (Xen toolstack only recognizes that and MFNs,
46 * no other fancy value).
48 * On lookup we spot that the entry points to p2m_identity and return the
49 * identity value instead of dereferencing and returning INVALID_P2M_ENTRY.
50 * If the entry points to an allocated page, we just proceed as before and
51 * return the PFN. If the PFN has IDENTITY_FRAME_BIT set we unmask that in
52 * appropriate functions (pfn_to_mfn).
54 * The reason for having the IDENTITY_FRAME_BIT instead of just returning the
55 * PFN is that we could find ourselves where pfn_to_mfn(pfn)==pfn for a
56 * non-identity pfn. To protect ourselves against we elect to set (and get) the
57 * IDENTITY_FRAME_BIT on all identity mapped PFNs.
59 * This simplistic diagram is used to explain the more subtle piece of code.
60 * There is also a digram of the P2M at the end that can help.
61 * Imagine your E820 looking as so:
64 * /-------------------+---------\/----\ /----------\ /---+-----\
65 * | System RAM | Sys RAM ||ACPI| | reserved | | Sys RAM |
66 * \-------------------+---------/\----/ \----------/ \---+-----/
69 * [1029MB = 263424 (0x40500), 2001MB = 512256 (0x7D100),
70 * 2048MB = 524288 (0x80000)]
72 * And dom0_mem=max:3GB,1GB is passed in to the guest, meaning memory past 1GB
73 * is actually not present (would have to kick the balloon driver to put it in).
75 * When we are told to set the PFNs for identity mapping (see patch: "xen/setup:
76 * Set identity mapping for non-RAM E820 and E820 gaps.") we pass in the start
77 * of the PFN and the end PFN (263424 and 512256 respectively). The first step
78 * is to reserve_brk a top leaf page if the p2m[1] is missing. The top leaf page
79 * covers 512^2 of page estate (1GB) and in case the start or end PFN is not
80 * aligned on 512^2*PAGE_SIZE (1GB) we loop on aligned 1GB PFNs from start pfn
81 * to end pfn. We reserve_brk top leaf pages if they are missing (means they
82 * point to p2m_mid_missing).
84 * With the E820 example above, 263424 is not 1GB aligned so we allocate a
85 * reserve_brk page which will cover the PFNs estate from 0x40000 to 0x80000.
86 * Each entry in the allocate page is "missing" (points to p2m_missing).
88 * Next stage is to determine if we need to do a more granular boundary check
89 * on the 4MB (or 2MB depending on architecture) off the start and end pfn's.
90 * We check if the start pfn and end pfn violate that boundary check, and if
91 * so reserve_brk a middle (p2m[x][y]) leaf page. This way we have a much finer
92 * granularity of setting which PFNs are missing and which ones are identity.
93 * In our example 263424 and 512256 both fail the check so we reserve_brk two
94 * pages. Populate them with INVALID_P2M_ENTRY (so they both have "missing"
95 * values) and assign them to p2m[1][2] and p2m[1][488] respectively.
97 * At this point we would at minimum reserve_brk one page, but could be up to
98 * three. Each call to set_phys_range_identity has at maximum a three page
99 * cost. If we were to query the P2M at this stage, all those entries from
100 * start PFN through end PFN (so 1029MB -> 2001MB) would return
101 * INVALID_P2M_ENTRY ("missing").
103 * The next step is to walk from the start pfn to the end pfn setting
104 * the IDENTITY_FRAME_BIT on each PFN. This is done in set_phys_range_identity.
105 * If we find that the middle leaf is pointing to p2m_missing we can swap it
106 * over to p2m_identity - this way covering 4MB (or 2MB) PFN space. At this
107 * point we do not need to worry about boundary aligment (so no need to
108 * reserve_brk a middle page, figure out which PFNs are "missing" and which
109 * ones are identity), as that has been done earlier. If we find that the
110 * middle leaf is not occupied by p2m_identity or p2m_missing, we dereference
111 * that page (which covers 512 PFNs) and set the appropriate PFN with
112 * IDENTITY_FRAME_BIT. In our example 263424 and 512256 end up there, and we
113 * set from p2m[1][2][256->511] and p2m[1][488][0->256] with
114 * IDENTITY_FRAME_BIT set.
116 * All other regions that are void (or not filled) either point to p2m_missing
117 * (considered missing) or have the default value of INVALID_P2M_ENTRY (also
118 * considered missing). In our case, p2m[1][2][0->255] and p2m[1][488][257->511]
119 * contain the INVALID_P2M_ENTRY value and are considered "missing."
121 * This is what the p2m ends up looking (for the E820 above) with this
124 * p2m /--------------\
125 * /-----\ | &mfn_list[0],| /-----------------\
126 * | 0 |------>| &mfn_list[1],| /---------------\ | ~0, ~0, .. |
127 * |-----| | ..., ~0, ~0 | | ~0, ~0, [x]---+----->| IDENTITY [@256] |
128 * | 1 |---\ \--------------/ | [p2m_identity]+\ | IDENTITY [@257] |
129 * |-----| \ | [p2m_identity]+\\ | .... |
130 * | 2 |--\ \-------------------->| ... | \\ \----------------/
131 * |-----| \ \---------------/ \\
132 * | 3 |\ \ \\ p2m_identity
133 * |-----| \ \-------------------->/---------------\ /-----------------\
134 * | .. +->+ | [p2m_identity]+-->| ~0, ~0, ~0, ... |
135 * \-----/ / | [p2m_identity]+-->| ..., ~0 |
136 * / /---------------\ | .... | \-----------------/
137 * / | IDENTITY[@0] | /-+-[x], ~0, ~0.. |
138 * / | IDENTITY[@256]|<----/ \---------------/
140 * | \---------------/
142 * p2m_mid_missing p2m_missing
143 * /-----------------\ /------------\
144 * | [p2m_missing] +---->| ~0, ~0, ~0 |
145 * | [p2m_missing] +---->| ..., ~0 |
146 * \-----------------/ \------------/
148 * where ~0 is INVALID_P2M_ENTRY. IDENTITY is (PFN | IDENTITY_BIT)
151 #include <linux/init.h>
152 #include <linux/module.h>
153 #include <linux/list.h>
154 #include <linux/hash.h>
155 #include <linux/sched.h>
156 #include <linux/seq_file.h>
158 #include <asm/cache.h>
159 #include <asm/setup.h>
161 #include <asm/xen/page.h>
162 #include <asm/xen/hypercall.h>
163 #include <asm/xen/hypervisor.h>
164 #include <xen/balloon.h>
165 #include <xen/grant_table.h>
167 #include "multicalls.h"
170 static void __init
m2p_override_init(void);
172 unsigned long xen_max_p2m_pfn __read_mostly
;
174 #define P2M_PER_PAGE (PAGE_SIZE / sizeof(unsigned long))
175 #define P2M_MID_PER_PAGE (PAGE_SIZE / sizeof(unsigned long *))
176 #define P2M_TOP_PER_PAGE (PAGE_SIZE / sizeof(unsigned long **))
178 #define MAX_P2M_PFN (P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE)
180 /* Placeholders for holes in the address space */
181 static RESERVE_BRK_ARRAY(unsigned long, p2m_missing
, P2M_PER_PAGE
);
182 static RESERVE_BRK_ARRAY(unsigned long *, p2m_mid_missing
, P2M_MID_PER_PAGE
);
183 static RESERVE_BRK_ARRAY(unsigned long, p2m_mid_missing_mfn
, P2M_MID_PER_PAGE
);
185 static RESERVE_BRK_ARRAY(unsigned long **, p2m_top
, P2M_TOP_PER_PAGE
);
186 static RESERVE_BRK_ARRAY(unsigned long, p2m_top_mfn
, P2M_TOP_PER_PAGE
);
187 static RESERVE_BRK_ARRAY(unsigned long *, p2m_top_mfn_p
, P2M_TOP_PER_PAGE
);
189 static RESERVE_BRK_ARRAY(unsigned long, p2m_identity
, P2M_PER_PAGE
);
191 RESERVE_BRK(p2m_mid
, PAGE_SIZE
* (MAX_DOMAIN_PAGES
/ (P2M_PER_PAGE
* P2M_MID_PER_PAGE
)));
192 RESERVE_BRK(p2m_mid_mfn
, PAGE_SIZE
* (MAX_DOMAIN_PAGES
/ (P2M_PER_PAGE
* P2M_MID_PER_PAGE
)));
194 /* We might hit two boundary violations at the start and end, at max each
195 * boundary violation will require three middle nodes. */
196 RESERVE_BRK(p2m_mid_identity
, PAGE_SIZE
* 2 * 3);
198 /* When we populate back during bootup, the amount of pages can vary. The
199 * max we have is seen is 395979, but that does not mean it can't be more.
200 * Some machines can have 3GB I/O holes even. With early_can_reuse_p2m_middle
201 * it can re-use Xen provided mfn_list array, so we only need to allocate at
202 * most three P2M top nodes. */
203 RESERVE_BRK(p2m_populated
, PAGE_SIZE
* 3);
205 static inline unsigned p2m_top_index(unsigned long pfn
)
207 BUG_ON(pfn
>= MAX_P2M_PFN
);
208 return pfn
/ (P2M_MID_PER_PAGE
* P2M_PER_PAGE
);
211 static inline unsigned p2m_mid_index(unsigned long pfn
)
213 return (pfn
/ P2M_PER_PAGE
) % P2M_MID_PER_PAGE
;
216 static inline unsigned p2m_index(unsigned long pfn
)
218 return pfn
% P2M_PER_PAGE
;
221 static void p2m_top_init(unsigned long ***top
)
225 for (i
= 0; i
< P2M_TOP_PER_PAGE
; i
++)
226 top
[i
] = p2m_mid_missing
;
229 static void p2m_top_mfn_init(unsigned long *top
)
233 for (i
= 0; i
< P2M_TOP_PER_PAGE
; i
++)
234 top
[i
] = virt_to_mfn(p2m_mid_missing_mfn
);
237 static void p2m_top_mfn_p_init(unsigned long **top
)
241 for (i
= 0; i
< P2M_TOP_PER_PAGE
; i
++)
242 top
[i
] = p2m_mid_missing_mfn
;
245 static void p2m_mid_init(unsigned long **mid
)
249 for (i
= 0; i
< P2M_MID_PER_PAGE
; i
++)
250 mid
[i
] = p2m_missing
;
253 static void p2m_mid_mfn_init(unsigned long *mid
)
257 for (i
= 0; i
< P2M_MID_PER_PAGE
; i
++)
258 mid
[i
] = virt_to_mfn(p2m_missing
);
261 static void p2m_init(unsigned long *p2m
)
265 for (i
= 0; i
< P2M_MID_PER_PAGE
; i
++)
266 p2m
[i
] = INVALID_P2M_ENTRY
;
270 * Build the parallel p2m_top_mfn and p2m_mid_mfn structures
272 * This is called both at boot time, and after resuming from suspend:
273 * - At boot time we're called very early, and must use extend_brk()
274 * to allocate memory.
276 * - After resume we're called from within stop_machine, but the mfn
277 * tree should alreay be completely allocated.
279 void __ref
xen_build_mfn_list_list(void)
283 /* Pre-initialize p2m_top_mfn to be completely missing */
284 if (p2m_top_mfn
== NULL
) {
285 p2m_mid_missing_mfn
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
286 p2m_mid_mfn_init(p2m_mid_missing_mfn
);
288 p2m_top_mfn_p
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
289 p2m_top_mfn_p_init(p2m_top_mfn_p
);
291 p2m_top_mfn
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
292 p2m_top_mfn_init(p2m_top_mfn
);
294 /* Reinitialise, mfn's all change after migration */
295 p2m_mid_mfn_init(p2m_mid_missing_mfn
);
298 for (pfn
= 0; pfn
< xen_max_p2m_pfn
; pfn
+= P2M_PER_PAGE
) {
299 unsigned topidx
= p2m_top_index(pfn
);
300 unsigned mididx
= p2m_mid_index(pfn
);
302 unsigned long *mid_mfn_p
;
304 mid
= p2m_top
[topidx
];
305 mid_mfn_p
= p2m_top_mfn_p
[topidx
];
307 /* Don't bother allocating any mfn mid levels if
308 * they're just missing, just update the stored mfn,
309 * since all could have changed over a migrate.
311 if (mid
== p2m_mid_missing
) {
313 BUG_ON(mid_mfn_p
!= p2m_mid_missing_mfn
);
314 p2m_top_mfn
[topidx
] = virt_to_mfn(p2m_mid_missing_mfn
);
315 pfn
+= (P2M_MID_PER_PAGE
- 1) * P2M_PER_PAGE
;
319 if (mid_mfn_p
== p2m_mid_missing_mfn
) {
321 * XXX boot-time only! We should never find
322 * missing parts of the mfn tree after
323 * runtime. extend_brk() will BUG if we call
326 mid_mfn_p
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
327 p2m_mid_mfn_init(mid_mfn_p
);
329 p2m_top_mfn_p
[topidx
] = mid_mfn_p
;
332 p2m_top_mfn
[topidx
] = virt_to_mfn(mid_mfn_p
);
333 mid_mfn_p
[mididx
] = virt_to_mfn(mid
[mididx
]);
337 void xen_setup_mfn_list_list(void)
339 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
341 HYPERVISOR_shared_info
->arch
.pfn_to_mfn_frame_list_list
=
342 virt_to_mfn(p2m_top_mfn
);
343 HYPERVISOR_shared_info
->arch
.max_pfn
= xen_max_p2m_pfn
;
346 /* Set up p2m_top to point to the domain-builder provided p2m pages */
347 void __init
xen_build_dynamic_phys_to_machine(void)
349 unsigned long *mfn_list
= (unsigned long *)xen_start_info
->mfn_list
;
350 unsigned long max_pfn
= min(MAX_DOMAIN_PAGES
, xen_start_info
->nr_pages
);
353 xen_max_p2m_pfn
= max_pfn
;
355 p2m_missing
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
356 p2m_init(p2m_missing
);
358 p2m_mid_missing
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
359 p2m_mid_init(p2m_mid_missing
);
361 p2m_top
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
362 p2m_top_init(p2m_top
);
364 p2m_identity
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
365 p2m_init(p2m_identity
);
368 * The domain builder gives us a pre-constructed p2m array in
369 * mfn_list for all the pages initially given to us, so we just
370 * need to graft that into our tree structure.
372 for (pfn
= 0; pfn
< max_pfn
; pfn
+= P2M_PER_PAGE
) {
373 unsigned topidx
= p2m_top_index(pfn
);
374 unsigned mididx
= p2m_mid_index(pfn
);
376 if (p2m_top
[topidx
] == p2m_mid_missing
) {
377 unsigned long **mid
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
380 p2m_top
[topidx
] = mid
;
384 * As long as the mfn_list has enough entries to completely
385 * fill a p2m page, pointing into the array is ok. But if
386 * not the entries beyond the last pfn will be undefined.
388 if (unlikely(pfn
+ P2M_PER_PAGE
> max_pfn
)) {
389 unsigned long p2midx
;
391 p2midx
= max_pfn
% P2M_PER_PAGE
;
392 for ( ; p2midx
< P2M_PER_PAGE
; p2midx
++)
393 mfn_list
[pfn
+ p2midx
] = INVALID_P2M_ENTRY
;
395 p2m_top
[topidx
][mididx
] = &mfn_list
[pfn
];
401 #include <linux/bootmem.h>
402 unsigned long __init
xen_revector_p2m_tree(void)
404 unsigned long va_start
;
405 unsigned long va_end
;
407 unsigned long pfn_free
= 0;
408 unsigned long *mfn_list
= NULL
;
411 va_start
= xen_start_info
->mfn_list
;
412 /*We copy in increments of P2M_PER_PAGE * sizeof(unsigned long),
413 * so make sure it is rounded up to that */
414 size
= PAGE_ALIGN(xen_start_info
->nr_pages
* sizeof(unsigned long));
415 va_end
= va_start
+ size
;
417 /* If we were revectored already, don't do it again. */
418 if (va_start
<= __START_KERNEL_map
&& va_start
>= __PAGE_OFFSET
)
421 mfn_list
= alloc_bootmem_align(size
, PAGE_SIZE
);
423 pr_warn("Could not allocate space for a new P2M tree!\n");
424 return xen_start_info
->mfn_list
;
426 /* Fill it out with INVALID_P2M_ENTRY value */
427 memset(mfn_list
, 0xFF, size
);
429 for (pfn
= 0; pfn
< ALIGN(MAX_DOMAIN_PAGES
, P2M_PER_PAGE
); pfn
+= P2M_PER_PAGE
) {
430 unsigned topidx
= p2m_top_index(pfn
);
432 unsigned long *mid_p
;
434 if (!p2m_top
[topidx
])
437 if (p2m_top
[topidx
] == p2m_mid_missing
)
440 mididx
= p2m_mid_index(pfn
);
441 mid_p
= p2m_top
[topidx
][mididx
];
444 if ((mid_p
== p2m_missing
) || (mid_p
== p2m_identity
))
447 if ((unsigned long)mid_p
== INVALID_P2M_ENTRY
)
450 /* The old va. Rebase it on mfn_list */
451 if (mid_p
>= (unsigned long *)va_start
&& mid_p
<= (unsigned long *)va_end
) {
454 if (pfn_free
> (size
/ sizeof(unsigned long))) {
455 WARN(1, "Only allocated for %ld pages, but we want %ld!\n",
456 size
/ sizeof(unsigned long), pfn_free
);
459 new = &mfn_list
[pfn_free
];
461 copy_page(new, mid_p
);
462 p2m_top
[topidx
][mididx
] = &mfn_list
[pfn_free
];
463 p2m_top_mfn_p
[topidx
][mididx
] = virt_to_mfn(&mfn_list
[pfn_free
]);
465 pfn_free
+= P2M_PER_PAGE
;
468 /* This should be the leafs allocated for identity from _brk. */
470 return (unsigned long)mfn_list
;
474 unsigned long __init
xen_revector_p2m_tree(void)
479 unsigned long get_phys_to_machine(unsigned long pfn
)
481 unsigned topidx
, mididx
, idx
;
483 if (unlikely(pfn
>= MAX_P2M_PFN
))
484 return INVALID_P2M_ENTRY
;
486 topidx
= p2m_top_index(pfn
);
487 mididx
= p2m_mid_index(pfn
);
488 idx
= p2m_index(pfn
);
491 * The INVALID_P2M_ENTRY is filled in both p2m_*identity
492 * and in p2m_*missing, so returning the INVALID_P2M_ENTRY
495 if (p2m_top
[topidx
][mididx
] == p2m_identity
)
496 return IDENTITY_FRAME(pfn
);
498 return p2m_top
[topidx
][mididx
][idx
];
500 EXPORT_SYMBOL_GPL(get_phys_to_machine
);
502 static void *alloc_p2m_page(void)
504 return (void *)__get_free_page(GFP_KERNEL
| __GFP_REPEAT
);
507 static void free_p2m_page(void *p
)
509 free_page((unsigned long)p
);
513 * Fully allocate the p2m structure for a given pfn. We need to check
514 * that both the top and mid levels are allocated, and make sure the
515 * parallel mfn tree is kept in sync. We may race with other cpus, so
516 * the new pages are installed with cmpxchg; if we lose the race then
517 * simply free the page we allocated and use the one that's there.
519 static bool alloc_p2m(unsigned long pfn
)
521 unsigned topidx
, mididx
;
522 unsigned long ***top_p
, **mid
;
523 unsigned long *top_mfn_p
, *mid_mfn
;
525 topidx
= p2m_top_index(pfn
);
526 mididx
= p2m_mid_index(pfn
);
528 top_p
= &p2m_top
[topidx
];
531 if (mid
== p2m_mid_missing
) {
532 /* Mid level is missing, allocate a new one */
533 mid
= alloc_p2m_page();
539 if (cmpxchg(top_p
, p2m_mid_missing
, mid
) != p2m_mid_missing
)
543 top_mfn_p
= &p2m_top_mfn
[topidx
];
544 mid_mfn
= p2m_top_mfn_p
[topidx
];
546 BUG_ON(virt_to_mfn(mid_mfn
) != *top_mfn_p
);
548 if (mid_mfn
== p2m_mid_missing_mfn
) {
549 /* Separately check the mid mfn level */
550 unsigned long missing_mfn
;
551 unsigned long mid_mfn_mfn
;
553 mid_mfn
= alloc_p2m_page();
557 p2m_mid_mfn_init(mid_mfn
);
559 missing_mfn
= virt_to_mfn(p2m_mid_missing_mfn
);
560 mid_mfn_mfn
= virt_to_mfn(mid_mfn
);
561 if (cmpxchg(top_mfn_p
, missing_mfn
, mid_mfn_mfn
) != missing_mfn
)
562 free_p2m_page(mid_mfn
);
564 p2m_top_mfn_p
[topidx
] = mid_mfn
;
567 if (p2m_top
[topidx
][mididx
] == p2m_identity
||
568 p2m_top
[topidx
][mididx
] == p2m_missing
) {
569 /* p2m leaf page is missing */
571 unsigned long *p2m_orig
= p2m_top
[topidx
][mididx
];
573 p2m
= alloc_p2m_page();
579 if (cmpxchg(&mid
[mididx
], p2m_orig
, p2m
) != p2m_orig
)
582 mid_mfn
[mididx
] = virt_to_mfn(p2m
);
588 static bool __init
early_alloc_p2m_middle(unsigned long pfn
, bool check_boundary
)
590 unsigned topidx
, mididx
, idx
;
592 unsigned long *mid_mfn_p
;
594 topidx
= p2m_top_index(pfn
);
595 mididx
= p2m_mid_index(pfn
);
596 idx
= p2m_index(pfn
);
598 /* Pfff.. No boundary cross-over, lets get out. */
599 if (!idx
&& check_boundary
)
602 WARN(p2m_top
[topidx
][mididx
] == p2m_identity
,
603 "P2M[%d][%d] == IDENTITY, should be MISSING (or alloced)!\n",
607 * Could be done by xen_build_dynamic_phys_to_machine..
609 if (p2m_top
[topidx
][mididx
] != p2m_missing
)
612 /* Boundary cross-over for the edges: */
613 p2m
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
617 p2m_top
[topidx
][mididx
] = p2m
;
619 /* For save/restore we need to MFN of the P2M saved */
621 mid_mfn_p
= p2m_top_mfn_p
[topidx
];
622 WARN(mid_mfn_p
[mididx
] != virt_to_mfn(p2m_missing
),
623 "P2M_TOP_P[%d][%d] != MFN of p2m_missing!\n",
625 mid_mfn_p
[mididx
] = virt_to_mfn(p2m
);
630 static bool __init
early_alloc_p2m(unsigned long pfn
)
632 unsigned topidx
= p2m_top_index(pfn
);
633 unsigned long *mid_mfn_p
;
636 mid
= p2m_top
[topidx
];
637 mid_mfn_p
= p2m_top_mfn_p
[topidx
];
638 if (mid
== p2m_mid_missing
) {
639 mid
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
643 p2m_top
[topidx
] = mid
;
645 BUG_ON(mid_mfn_p
!= p2m_mid_missing_mfn
);
647 /* And the save/restore P2M tables.. */
648 if (mid_mfn_p
== p2m_mid_missing_mfn
) {
649 mid_mfn_p
= extend_brk(PAGE_SIZE
, PAGE_SIZE
);
650 p2m_mid_mfn_init(mid_mfn_p
);
652 p2m_top_mfn_p
[topidx
] = mid_mfn_p
;
653 p2m_top_mfn
[topidx
] = virt_to_mfn(mid_mfn_p
);
654 /* Note: we don't set mid_mfn_p[midix] here,
655 * look in early_alloc_p2m_middle */
661 * Skim over the P2M tree looking at pages that are either filled with
662 * INVALID_P2M_ENTRY or with 1:1 PFNs. If found, re-use that page and
663 * replace the P2M leaf with a p2m_missing or p2m_identity.
664 * Stick the old page in the new P2M tree location.
666 bool __init
early_can_reuse_p2m_middle(unsigned long set_pfn
, unsigned long set_mfn
)
673 unsigned long *mid_mfn_p
;
677 /* We only look when this entails a P2M middle layer */
678 if (p2m_index(set_pfn
))
681 for (pfn
= 0; pfn
< MAX_DOMAIN_PAGES
; pfn
+= P2M_PER_PAGE
) {
682 topidx
= p2m_top_index(pfn
);
684 if (!p2m_top
[topidx
])
687 if (p2m_top
[topidx
] == p2m_mid_missing
)
690 mididx
= p2m_mid_index(pfn
);
691 p2m
= p2m_top
[topidx
][mididx
];
695 if ((p2m
== p2m_missing
) || (p2m
== p2m_identity
))
698 if ((unsigned long)p2m
== INVALID_P2M_ENTRY
)
703 for (idx
= 0; idx
< P2M_PER_PAGE
; idx
++) {
704 /* IDENTITY_PFNs are 1:1 */
705 if (p2m
[idx
] == IDENTITY_FRAME(pfn
+ idx
))
707 else if (p2m
[idx
] == INVALID_P2M_ENTRY
)
712 if ((ident_pfns
== P2M_PER_PAGE
) || (inv_pfns
== P2M_PER_PAGE
))
717 /* Found one, replace old with p2m_identity or p2m_missing */
718 p2m_top
[topidx
][mididx
] = (ident_pfns
? p2m_identity
: p2m_missing
);
719 /* And the other for save/restore.. */
720 mid_mfn_p
= p2m_top_mfn_p
[topidx
];
721 /* NOTE: Even if it is a p2m_identity it should still be point to
722 * a page filled with INVALID_P2M_ENTRY entries. */
723 mid_mfn_p
[mididx
] = virt_to_mfn(p2m_missing
);
725 /* Reset where we want to stick the old page in. */
726 topidx
= p2m_top_index(set_pfn
);
727 mididx
= p2m_mid_index(set_pfn
);
729 /* This shouldn't happen */
730 if (WARN_ON(p2m_top
[topidx
] == p2m_mid_missing
))
731 early_alloc_p2m(set_pfn
);
733 if (WARN_ON(p2m_top
[topidx
][mididx
] != p2m_missing
))
737 p2m_top
[topidx
][mididx
] = p2m
;
738 mid_mfn_p
= p2m_top_mfn_p
[topidx
];
739 mid_mfn_p
[mididx
] = virt_to_mfn(p2m
);
743 bool __init
early_set_phys_to_machine(unsigned long pfn
, unsigned long mfn
)
745 if (unlikely(!__set_phys_to_machine(pfn
, mfn
))) {
746 if (!early_alloc_p2m(pfn
))
749 if (early_can_reuse_p2m_middle(pfn
, mfn
))
750 return __set_phys_to_machine(pfn
, mfn
);
752 if (!early_alloc_p2m_middle(pfn
, false /* boundary crossover OK!*/))
755 if (!__set_phys_to_machine(pfn
, mfn
))
761 unsigned long __init
set_phys_range_identity(unsigned long pfn_s
,
766 if (unlikely(pfn_s
>= MAX_P2M_PFN
|| pfn_e
>= MAX_P2M_PFN
))
769 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap
)))
770 return pfn_e
- pfn_s
;
775 for (pfn
= (pfn_s
& ~(P2M_MID_PER_PAGE
* P2M_PER_PAGE
- 1));
776 pfn
< ALIGN(pfn_e
, (P2M_MID_PER_PAGE
* P2M_PER_PAGE
));
777 pfn
+= P2M_MID_PER_PAGE
* P2M_PER_PAGE
)
779 WARN_ON(!early_alloc_p2m(pfn
));
782 early_alloc_p2m_middle(pfn_s
, true);
783 early_alloc_p2m_middle(pfn_e
, true);
785 for (pfn
= pfn_s
; pfn
< pfn_e
; pfn
++)
786 if (!__set_phys_to_machine(pfn
, IDENTITY_FRAME(pfn
)))
789 if (!WARN((pfn
- pfn_s
) != (pfn_e
- pfn_s
),
790 "Identity mapping failed. We are %ld short of 1-1 mappings!\n",
791 (pfn_e
- pfn_s
) - (pfn
- pfn_s
)))
792 printk(KERN_DEBUG
"1-1 mapping on %lx->%lx\n", pfn_s
, pfn
);
797 /* Try to install p2m mapping; fail if intermediate bits missing */
798 bool __set_phys_to_machine(unsigned long pfn
, unsigned long mfn
)
800 unsigned topidx
, mididx
, idx
;
802 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap
))) {
803 BUG_ON(pfn
!= mfn
&& mfn
!= INVALID_P2M_ENTRY
);
806 if (unlikely(pfn
>= MAX_P2M_PFN
)) {
807 BUG_ON(mfn
!= INVALID_P2M_ENTRY
);
811 topidx
= p2m_top_index(pfn
);
812 mididx
= p2m_mid_index(pfn
);
813 idx
= p2m_index(pfn
);
815 /* For sparse holes were the p2m leaf has real PFN along with
816 * PCI holes, stick in the PFN as the MFN value.
818 if (mfn
!= INVALID_P2M_ENTRY
&& (mfn
& IDENTITY_FRAME_BIT
)) {
819 if (p2m_top
[topidx
][mididx
] == p2m_identity
)
822 /* Swap over from MISSING to IDENTITY if needed. */
823 if (p2m_top
[topidx
][mididx
] == p2m_missing
) {
824 WARN_ON(cmpxchg(&p2m_top
[topidx
][mididx
], p2m_missing
,
825 p2m_identity
) != p2m_missing
);
830 if (p2m_top
[topidx
][mididx
] == p2m_missing
)
831 return mfn
== INVALID_P2M_ENTRY
;
833 p2m_top
[topidx
][mididx
][idx
] = mfn
;
838 bool set_phys_to_machine(unsigned long pfn
, unsigned long mfn
)
840 if (unlikely(!__set_phys_to_machine(pfn
, mfn
))) {
844 if (!__set_phys_to_machine(pfn
, mfn
))
851 #define M2P_OVERRIDE_HASH_SHIFT 10
852 #define M2P_OVERRIDE_HASH (1 << M2P_OVERRIDE_HASH_SHIFT)
854 static RESERVE_BRK_ARRAY(struct list_head
, m2p_overrides
, M2P_OVERRIDE_HASH
);
855 static DEFINE_SPINLOCK(m2p_override_lock
);
857 static void __init
m2p_override_init(void)
861 m2p_overrides
= extend_brk(sizeof(*m2p_overrides
) * M2P_OVERRIDE_HASH
,
862 sizeof(unsigned long));
864 for (i
= 0; i
< M2P_OVERRIDE_HASH
; i
++)
865 INIT_LIST_HEAD(&m2p_overrides
[i
]);
868 static unsigned long mfn_hash(unsigned long mfn
)
870 return hash_long(mfn
, M2P_OVERRIDE_HASH_SHIFT
);
873 /* Add an MFN override for a particular page */
874 int m2p_add_override(unsigned long mfn
, struct page
*page
,
875 struct gnttab_map_grant_ref
*kmap_op
)
879 unsigned long uninitialized_var(address
);
883 pfn
= page_to_pfn(page
);
884 if (!PageHighMem(page
)) {
885 address
= (unsigned long)__va(pfn
<< PAGE_SHIFT
);
886 ptep
= lookup_address(address
, &level
);
887 if (WARN(ptep
== NULL
|| level
!= PG_LEVEL_4K
,
888 "m2p_add_override: pfn %lx not mapped", pfn
))
891 WARN_ON(PagePrivate(page
));
892 SetPagePrivate(page
);
893 set_page_private(page
, mfn
);
894 page
->index
= pfn_to_mfn(pfn
);
896 if (unlikely(!set_phys_to_machine(pfn
, FOREIGN_FRAME(mfn
))))
899 if (kmap_op
!= NULL
) {
900 if (!PageHighMem(page
)) {
901 struct multicall_space mcs
=
902 xen_mc_entry(sizeof(*kmap_op
));
904 MULTI_grant_table_op(mcs
.mc
,
905 GNTTABOP_map_grant_ref
, kmap_op
, 1);
907 xen_mc_issue(PARAVIRT_LAZY_MMU
);
910 spin_lock_irqsave(&m2p_override_lock
, flags
);
911 list_add(&page
->lru
, &m2p_overrides
[mfn_hash(mfn
)]);
912 spin_unlock_irqrestore(&m2p_override_lock
, flags
);
914 /* p2m(m2p(mfn)) == mfn: the mfn is already present somewhere in
915 * this domain. Set the FOREIGN_FRAME_BIT in the p2m for the other
916 * pfn so that the following mfn_to_pfn(mfn) calls will return the
917 * pfn from the m2p_override (the backend pfn) instead.
918 * We need to do this because the pages shared by the frontend
919 * (xen-blkfront) can be already locked (lock_page, called by
920 * do_read_cache_page); when the userspace backend tries to use them
921 * with direct_IO, mfn_to_pfn returns the pfn of the frontend, so
922 * do_blockdev_direct_IO is going to try to lock the same pages
923 * again resulting in a deadlock.
924 * As a side effect get_user_pages_fast might not be safe on the
925 * frontend pages while they are being shared with the backend,
926 * because mfn_to_pfn (that ends up being called by GUPF) will
927 * return the backend pfn rather than the frontend pfn. */
928 pfn
= mfn_to_pfn_no_overrides(mfn
);
929 if (get_phys_to_machine(pfn
) == mfn
)
930 set_phys_to_machine(pfn
, FOREIGN_FRAME(mfn
));
934 EXPORT_SYMBOL_GPL(m2p_add_override
);
935 int m2p_remove_override(struct page
*page
,
936 struct gnttab_map_grant_ref
*kmap_op
)
941 unsigned long uninitialized_var(address
);
945 pfn
= page_to_pfn(page
);
946 mfn
= get_phys_to_machine(pfn
);
947 if (mfn
== INVALID_P2M_ENTRY
|| !(mfn
& FOREIGN_FRAME_BIT
))
950 if (!PageHighMem(page
)) {
951 address
= (unsigned long)__va(pfn
<< PAGE_SHIFT
);
952 ptep
= lookup_address(address
, &level
);
954 if (WARN(ptep
== NULL
|| level
!= PG_LEVEL_4K
,
955 "m2p_remove_override: pfn %lx not mapped", pfn
))
959 spin_lock_irqsave(&m2p_override_lock
, flags
);
960 list_del(&page
->lru
);
961 spin_unlock_irqrestore(&m2p_override_lock
, flags
);
962 WARN_ON(!PagePrivate(page
));
963 ClearPagePrivate(page
);
965 set_phys_to_machine(pfn
, page
->index
);
966 if (kmap_op
!= NULL
) {
967 if (!PageHighMem(page
)) {
968 struct multicall_space mcs
;
969 struct gnttab_unmap_and_replace
*unmap_op
;
970 struct page
*scratch_page
= get_balloon_scratch_page();
971 unsigned long scratch_page_address
= (unsigned long)
972 __va(page_to_pfn(scratch_page
) << PAGE_SHIFT
);
975 * It might be that we queued all the m2p grant table
976 * hypercalls in a multicall, then m2p_remove_override
977 * get called before the multicall has actually been
978 * issued. In this case handle is going to -1 because
979 * it hasn't been modified yet.
981 if (kmap_op
->handle
== -1)
984 * Now if kmap_op->handle is negative it means that the
985 * hypercall actually returned an error.
987 if (kmap_op
->handle
== GNTST_general_error
) {
988 printk(KERN_WARNING
"m2p_remove_override: "
989 "pfn %lx mfn %lx, failed to modify kernel mappings",
991 put_balloon_scratch_page();
997 mcs
= __xen_mc_entry(
998 sizeof(struct gnttab_unmap_and_replace
));
1000 unmap_op
->host_addr
= kmap_op
->host_addr
;
1001 unmap_op
->new_addr
= scratch_page_address
;
1002 unmap_op
->handle
= kmap_op
->handle
;
1004 MULTI_grant_table_op(mcs
.mc
,
1005 GNTTABOP_unmap_and_replace
, unmap_op
, 1);
1007 mcs
= __xen_mc_entry(0);
1008 MULTI_update_va_mapping(mcs
.mc
, scratch_page_address
,
1009 pfn_pte(page_to_pfn(scratch_page
),
1010 PAGE_KERNEL_RO
), 0);
1012 xen_mc_issue(PARAVIRT_LAZY_MMU
);
1014 kmap_op
->host_addr
= 0;
1015 put_balloon_scratch_page();
1019 /* p2m(m2p(mfn)) == FOREIGN_FRAME(mfn): the mfn is already present
1020 * somewhere in this domain, even before being added to the
1021 * m2p_override (see comment above in m2p_add_override).
1022 * If there are no other entries in the m2p_override corresponding
1023 * to this mfn, then remove the FOREIGN_FRAME_BIT from the p2m for
1024 * the original pfn (the one shared by the frontend): the backend
1025 * cannot do any IO on this page anymore because it has been
1026 * unshared. Removing the FOREIGN_FRAME_BIT from the p2m entry of
1027 * the original pfn causes mfn_to_pfn(mfn) to return the frontend
1029 mfn
&= ~FOREIGN_FRAME_BIT
;
1030 pfn
= mfn_to_pfn_no_overrides(mfn
);
1031 if (get_phys_to_machine(pfn
) == FOREIGN_FRAME(mfn
) &&
1032 m2p_find_override(mfn
) == NULL
)
1033 set_phys_to_machine(pfn
, mfn
);
1037 EXPORT_SYMBOL_GPL(m2p_remove_override
);
1039 struct page
*m2p_find_override(unsigned long mfn
)
1041 unsigned long flags
;
1042 struct list_head
*bucket
= &m2p_overrides
[mfn_hash(mfn
)];
1043 struct page
*p
, *ret
;
1047 spin_lock_irqsave(&m2p_override_lock
, flags
);
1049 list_for_each_entry(p
, bucket
, lru
) {
1050 if (page_private(p
) == mfn
) {
1056 spin_unlock_irqrestore(&m2p_override_lock
, flags
);
1061 unsigned long m2p_find_override_pfn(unsigned long mfn
, unsigned long pfn
)
1063 struct page
*p
= m2p_find_override(mfn
);
1064 unsigned long ret
= pfn
;
1067 ret
= page_to_pfn(p
);
1071 EXPORT_SYMBOL_GPL(m2p_find_override_pfn
);
1073 #ifdef CONFIG_XEN_DEBUG_FS
1074 #include <linux/debugfs.h>
1075 #include "debugfs.h"
1076 static int p2m_dump_show(struct seq_file
*m
, void *v
)
1078 static const char * const level_name
[] = { "top", "middle",
1079 "entry", "abnormal", "error"};
1080 #define TYPE_IDENTITY 0
1081 #define TYPE_MISSING 1
1083 #define TYPE_UNKNOWN 3
1084 static const char * const type_name
[] = {
1085 [TYPE_IDENTITY
] = "identity",
1086 [TYPE_MISSING
] = "missing",
1088 [TYPE_UNKNOWN
] = "abnormal"};
1089 unsigned long pfn
, prev_pfn_type
= 0, prev_pfn_level
= 0;
1090 unsigned int uninitialized_var(prev_level
);
1091 unsigned int uninitialized_var(prev_type
);
1096 for (pfn
= 0; pfn
< MAX_DOMAIN_PAGES
; pfn
++) {
1097 unsigned topidx
= p2m_top_index(pfn
);
1098 unsigned mididx
= p2m_mid_index(pfn
);
1099 unsigned idx
= p2m_index(pfn
);
1103 type
= TYPE_UNKNOWN
;
1104 if (p2m_top
[topidx
] == p2m_mid_missing
) {
1105 lvl
= 0; type
= TYPE_MISSING
;
1106 } else if (p2m_top
[topidx
] == NULL
) {
1107 lvl
= 0; type
= TYPE_UNKNOWN
;
1108 } else if (p2m_top
[topidx
][mididx
] == NULL
) {
1109 lvl
= 1; type
= TYPE_UNKNOWN
;
1110 } else if (p2m_top
[topidx
][mididx
] == p2m_identity
) {
1111 lvl
= 1; type
= TYPE_IDENTITY
;
1112 } else if (p2m_top
[topidx
][mididx
] == p2m_missing
) {
1113 lvl
= 1; type
= TYPE_MISSING
;
1114 } else if (p2m_top
[topidx
][mididx
][idx
] == 0) {
1115 lvl
= 2; type
= TYPE_UNKNOWN
;
1116 } else if (p2m_top
[topidx
][mididx
][idx
] == IDENTITY_FRAME(pfn
)) {
1117 lvl
= 2; type
= TYPE_IDENTITY
;
1118 } else if (p2m_top
[topidx
][mididx
][idx
] == INVALID_P2M_ENTRY
) {
1119 lvl
= 2; type
= TYPE_MISSING
;
1120 } else if (p2m_top
[topidx
][mididx
][idx
] == pfn
) {
1121 lvl
= 2; type
= TYPE_PFN
;
1122 } else if (p2m_top
[topidx
][mididx
][idx
] != pfn
) {
1123 lvl
= 2; type
= TYPE_PFN
;
1129 if (pfn
== MAX_DOMAIN_PAGES
-1) {
1131 type
= TYPE_UNKNOWN
;
1133 if (prev_type
!= type
) {
1134 seq_printf(m
, " [0x%lx->0x%lx] %s\n",
1135 prev_pfn_type
, pfn
, type_name
[prev_type
]);
1136 prev_pfn_type
= pfn
;
1139 if (prev_level
!= lvl
) {
1140 seq_printf(m
, " [0x%lx->0x%lx] level %s\n",
1141 prev_pfn_level
, pfn
, level_name
[prev_level
]);
1142 prev_pfn_level
= pfn
;
1147 #undef TYPE_IDENTITY
1153 static int p2m_dump_open(struct inode
*inode
, struct file
*filp
)
1155 return single_open(filp
, p2m_dump_show
, NULL
);
1158 static const struct file_operations p2m_dump_fops
= {
1159 .open
= p2m_dump_open
,
1161 .llseek
= seq_lseek
,
1162 .release
= single_release
,
1165 static struct dentry
*d_mmu_debug
;
1167 static int __init
xen_p2m_debugfs(void)
1169 struct dentry
*d_xen
= xen_init_debugfs();
1174 d_mmu_debug
= debugfs_create_dir("mmu", d_xen
);
1176 debugfs_create_file("p2m", 0600, d_mmu_debug
, NULL
, &p2m_dump_fops
);
1179 fs_initcall(xen_p2m_debugfs
);
1180 #endif /* CONFIG_XEN_DEBUG_FS */